The homogeneity marker serves to identify polygons (FOIs) on which several crops are being grown. It identifies heterogeneities on the scale of the FOIs declared primary crop's season and is not applicable to individual observations. That is the case, because a FOI could be heterogeneous in individual observations for many reasons. Some of the most common short-period heterogeneities are the consequences of:
- farming practices like partial mowing, which is especially common on large meadow polygons.
- intrinsic properties of the polygons, like intra-FOI differences of soil quality, moisture or many others. Fast-growing crops are especially susceptible to such heterogeneities.
To eliminate the influence of such short-term heterogeneities, we evaluate heterogeneity of a FOI on a season time-frame by aggregating relevant signals into season-spanning features.
The most relevant signals are:
- standard deviations and percentiles over the FOI's pixels for various signals. These can be bands like B02, B03 or indices such as NDVI and NBSI.
- we have also developed land-cover classifiers that can be used to roughly determine the ratio of a FOI's pixels that belong to a certain land-cover for each observation. Examples of this are the bare-soil (left) and the built-up pixel-masks (right). Red- and blue-colored pixels represent bare-soil and built-up land-cover respectively. As can be seen, for some types of heterogeneities, FOIs are expected to have pixels that are consistently different in terms of their predicted land-cover class.
|FOI = 6550882001||FOI = 6736785001|
The signals like the one listed above are aggregated into multi-month features for each signal inside a desired time-frame. For Slovenia, the most important averaging time-frame that has been identified is the 3-month averaging interval from 1st of May to 1st of August, since it coincides with the growing season of most crops. Below we show the distribution of
NDVI_std_3M_MAY, which represents the feature we get by performing a three-month average of the
NDVI_std from 1st of May to 1st of August. FOIs that are known to be homogeneous or heterogeneous are shown in orange and blue respectively.
Clearly, heterogeneous FOIs have higher
NDVI_std_3M_MAY (similar for other features), which allows us to distinguish between homogeneous and heterogeneous FOIs. The homogeneity marker uses many such engineered features to assign a probability of a FOI being homogeneous. If the probability is larger than 35 (on a scale of 0-100) we classify the FOI as homogeneous, while for probabilities lower than 35 a heterogeneous classification is assigned.